While there are still some, who doubt that climate change is going to happen, many others are trying to find answers to the question how to adapt to that change.
One major problem when trying to assess the impacts of climate change on water resources, is the large biases that are still present in the current climate models.
One way to overcome this, is to correct the model predictions by statistical means using past observation data. We used quantile-quantile transformations to both correct climate scenario output and increase the spatial resolution of the predictions. The cumulative distribution functions needed for this transformation were obtained non-parametrically by kernel smoothing. The optimal kernel width was determined by a maximum likelihood method.
The transformations were applied to precipitation from the three IPCC emission scenarios A1B, A2 and B1, calculated by two climate models and corrected using three precipitation products based on monthly observations, for the Nile Equatorial Lakes region in equatorial East Africa. Major results were that the large differences between the two models could effectively be reduced, while an, albeit small, climate change signal could be retained.
All codes for this project were implemented in Python and numpy. Using PyTables, it was possible to use the HDF5-format for data storage both easily and effectively. The newly available mapping features of the python interface in ArcGIS 10, enabled us to automatically create high quality maps of the spatial distribution of the precipitation and change signals for all datasets. Finally, the NetCDF4 package made it possible to easily transform the results into datasets, which can be readily used in tools like GrADS for further analysis.